Semantic segmentation of railway track images with deep convolutional neural networks

The condition of railway tracks needs to be periodically monitored to ensure passenger safety. Cameras mounted on a moving vehicle such as a hi-rail vehicle or geometry inspection car can generate large volumes of high resolution images. Extracting accurate information from those images has been challenging due to the clutter in the railroad environment. In this paper we describe a novel approach to visual track inspection using semantic segmentation with Deep Convolutional Neural Networks. We show that DCNNs trained end-to-end for material classification are more accurate than shallow learning machines with hand-engineered features and are more robust to noise. Our approach results in a material classification accuracy of 93.35% using 10 classes of materials. This allows for the detection of crumbling and chipped tie conditions at detection rates of 86.06% and 92.11%, respectively, at a false positive rate of 10 FP/mile on the 85-mile Northeast Corridor (NEC) 2012-2013 concrete tie dataset. Index Terms— Semantic Segmentation, Deep Convolutional Neural Networks, Railway Track Inspection, Material Classification. 1. INTRODUCTION Railway tracks need to be regularly inspected to ensure train safety. Crossties, also known as sleepers, are responsible for supporting the rails and maintaing track geometry within safety ranges. Tracks have been historically built with timber ties, but during the last half century, steel reinforced concrete has been the preferred material for building crossties. Concrete ties have several advantages over wood ties, such being a more uniform product, with better control of tolerances, as well as being well adapted for elastic fasteners, which control longitudinal forces better than conventional ones. Moreover, by being heavier than timber ties, concrete ties promote better track stability [1]. For all these reasons, concrete ties have been widely adopted, specially in high speed corridors. Although concrete ties have life expectancies of up to 50 years, they may fail prematurely for a variety of reasons, such as the result of alkali-silicone reaction (ASR) [2] or delayed ettringite formation (DEF) [3]. Ties may also develop fatigue cracks due to normal traffic or by being impacted by flying debris or track maintenance machinery. Once small cracks develop, repeated cycles of freezing and thawing will eventually lead to a bigger defects. This work was supported by the Federal Railroad Administration under contract DTFR53-13-C-00032. The authors thak Amtrak, ENSCO, Inc. and the Federal Railroad Administration for providing the data used in this paper. Le# Rail Right Rail